Lot sizing that holds up: test it on trades it has never seen

Most traders pick their lot size the same way. They try a few sizing rules on their history (a fixed lot, a percentage of balance, a multiple of the tested size) and keep the one that gives the most growth at a drawdown they can live with.

It feels like research. Most of the time it is curve fitting, just aimed at the lot size instead of the entry rules.

Why sizing overfits so easily

A sizing rule doesn't change which trades you take. It changes how much each one weighs. That sounds harmless, but sizing reacts to the order of your wins and losses:

• a compounding rule that happened to grow just before a lucky streak looks brilliant;
• the same rule a month later, just before a losing streak, looks reckless.

Then there is the number of tries. Take six sizing rules with eight values each. That is 48 settings. Even if none of them is truly better than your current sizing, the one that comes out on top of 48 noisy results will beat it on the history you tested, simply because it is the highest of 48 draws. Choosing it and judging it on the same trades guarantees a good-looking number. It tells you nothing about the trades to come.

The honest test: choose on one part, judge on another

The fix is old and simple. Split your history in two:

1. The earlier part is where you search. Try every rule and value you like there, and keep the few that lead.
2. The later part is where you judge, and only there. The settings never saw these trades while they were chosen.

A few details decide whether the test means anything:

• The judged part must be long enough. A few weeks of trades can make anything look good or bad. As a rule of thumb, ask for at least three months and enough separate trade groups. For a grid or basket EA, overlapping positions on one symbol behave like one bet, so count them as one.
• Compare like with like. More leverage always grows faster until it doesn't. Scale your current sizing to the same drawdown as the candidate before comparing growth; otherwise you are only measuring risk appetite.
• Judge under worse costs too. A sizing that only wins when spreads and slippage are kind is fragile. Rerun the judged part with doubled slippage.
• Look at the neighbours. If 1.5x holds up but 1.25x and 1.75x both collapse, you found a spike in the noise, not a setting.

Signs that a sizing does not hold up

• Growth on the judged part far below the search part, after scaling to the same drawdown.
• A drawdown on the judged part well beyond what the search part showed for periods of the same length.
• A result that disappears under stressed costs.
• A setting that only works at one exact value.

When nothing holds up

This is the most common outcome, and the most useful one. It means your current sizing is not the problem, and changing it would only add risk. No sizing rule can turn a strategy without an edge into one with an edge. It can only make the ride bigger or smaller. If the strategy itself loses on the part you judged, the honest next step is to look at the strategy, not at the lot size.

A short checklist before you change your lot size

1. Did you choose the setting on one part of your history and judge it on a later part only?
2. Is the judged part at least three months, with enough separate trade groups?
3. Did you compare against your current sizing scaled to the same drawdown?
4. Does it still hold up with doubled slippage?
5. Do the neighbouring values agree?

If any answer is no, the result is not ready to trust with real money.

How Strategy Ledger Pro runs this test

The SIZING OPTIMIZER in Strategy Ledger Pro's What-if does exactly this, on a backtest report or on your live account's closed trades:

• it tries the sizing rules (FIXED LOT, PER BALANCE, x SIZE, EQUAL RISK, EQUAL DD and RISK BUDGET, plus DAILY STOP if you switch it on) on an earlier part of your history;
• it checks the leading settings on a later part they have not seen;
• it compares each one with your own sizing at the same drawdown;
• it gives each a plain verdict: HOLDS UP, WEAKER, ONLY ON CHEAP COSTS or DOES NOT HOLD UP.

When nothing holds up, it says so first. The release notes publish how often it is fooled: on purely random test data with deposits and withdrawals in the history, it gave HOLDS UP in about 4.5% of runs, against a limit of 10% it is held to. Like the rest of the panel, it is read-only: it never places, changes or closes a trade.

Step-by-step guide to What-if: https://www.mql5.com/en/blogs/post/777484
Product page: https://www.mql5.com/en/market/product/192540